Continually evaluating large generative models provides a unique challenge. Often, human annotations are necessary to evaluate high-level properties of these models (e.g. in text or images). However, collecting human annotations of samples can be resource intensive, and using other machine learning systems to provide the annotations, or automatic evaluation, can introduce systematic errors into the evaluation. The Prediction Powered Inference (PPI) framework provides a way of leveraging both the statistical power of automatic evaluation and a small pool of labelled data to produce a low-variance, unbiased estimate of the quantity being evaluated for. However, most work on PPI considers a relatively sizable set of labelled samples, which is not always practical to obtain. To this end, we present two new PPI-based techniques that leverage robust regressors to produce even lower variance estimators in the few-label regime.

Large language models are trained on massive scrapes of the web, as required by current scaling laws. Most progress is made for English, given its abundance of high-quality pretraining data. For most other languages, however, such high quality pretraining data is unavailable. In this work, we study how to boost pretrained model performance in a data constrained target language by enlisting data from an auxiliary language for which high quality data is available. We study this by quantifying the performance gap between training with data in a data-rich auxiliary language compared with training in the target language, exploring the benefits of translation systems, studying the limitations of model scaling for data constrained languages, and proposing new methods for upsampling data from the auxiliary language. Our results show that stronger auxiliary datasets result in performance gains without modification to the model or training objective for close languages, and, in particular, that performance gains due to the development of more information-rich English pretraining datasets can extend to targeted language settings with limited data.

Google just announced that it has started equipping its online document editing tool Google Docs with new AI features. Google's own AI tool, called Gemini alongside OpenAI's ChatGPT and Microsoft's Copilot, has been able to generate images for a while now, but soon you'll be able to use Gemini to create AI-generated images directly within Google Docs itself. If you frequently use Google Docs, whether for personal use or as part of a larger team, and if you often need images and illustrations to support your text, this feature will boost your productivity. When you want to insert an AI-generated image into Google Docs, simply go to the Insert menu, then select Cover image Help me create an image. You'll then be able to describe what you want the image to be and choose from several different visual styles and formats.

Late last month, the US Treasury Department finalized new restrictions limiting what kinds of Chinese tech startups US venture capital firms can invest in for national security reasons. When they go into effect in January, the long-awaited measures will stop American VCs and other investors from pouring money into cutting-edge Chinese AI models. After president-elect Trump takes office a few weeks later, his administration may expand the rules and make them even tougher. While the US still leads the world in advanced AI development, the American government has grown increasingly concerned about China catching up soon. The new outbound investment restrictions are designed to work alongside other measures, such as export controls on advanced computer chips and the Committee on Foreign Investment in the United States (CFIUS), to collectively hamper--or at least slow down--the progress of Chinese AI companies.

Despite the remarkable capabilities of modern large language models (LLMs), the mechanisms behind their problem-solving abilities remain elusive. In this work, we aim to better understand how the learning dynamics of LLM finetuning shapes downstream generalization. Our analysis focuses on reasoning tasks, whose problem structure allows us to distinguish between memorization (the exact replication of reasoning steps from the training data) and performance (the correctness of the final solution). We find that a model's generalization behavior can be effectively characterized by a training metric we call pre-memorization train accuracy: the accuracy of model samples on training queries before they begin to copy the exact reasoning steps from the training set. On the dataset level, this metric is able to reliably predict test accuracy, achieving $R^2$ of around or exceeding 0.9 across various models (Llama3 8, Gemma2 9B), datasets (GSM8k, MATH), and training configurations. On a per-example level, this metric is also indicative of whether individual model predictions are robust to perturbations in the training query. By connecting a model's learning behavior to its generalization, pre-memorization train accuracy can guide targeted improvements to training strategies. We focus on data curation as an example, and show that prioritizing examples with low pre-memorization accuracy leads to 1.5-2x improvements in data efficiency compared to i.i.d. data scaling, and outperforms other standard data curation techniques.

Deep learning models have shown strong performance in load forecasting, but they generally require large amounts of data for model training before being applied to new scenarios, which limits their effectiveness in data-scarce scenarios. Inspired by the great success of pre-trained language models (LLMs) in natural language processing, this paper proposes a zero-shot load forecasting approach using an advanced LLM framework denoted as the Chronos model. By utilizing its extensive pre-trained knowledge, the Chronos model enables accurate load forecasting in data-scarce scenarios without the need for extensive data-specific training. Simulation results across five real-world datasets demonstrate that the Chronos model significantly outperforms nine popular baseline models for both deterministic and probabilistic load forecasting with various forecast horizons (e.g., 1 to 48 hours), even though the Chronos model is neither tailored nor fine-tuned to these specific load datasets. Notably, Chronos reduces root mean squared error (RMSE), continuous ranked probability score (CRPS), and quantile score (QS) by approximately 7.34%-84.30%, 19.63%-60.06%, and 22.83%-54.49%, respectively, compared to baseline models. These results highlight the superiority and flexibility of the Chronos model, positioning it as an effective solution in data-scarce scenarios.

Computer programming (coding) is indispensable for researchers across disciplines, yet it remains challenging to learn and time-consuming to carry out. Generative AI, particularly large language models (LLMs), has the potential to transform coding into intuitive conversations, but best practices and effective workflows are only emerging. We dissect AI-based coding through three key lenses: the nature and role of LLMs in coding (why), six types of coding assistance they provide (what), and a five-step workflow in action with practical implementation strategies (how). Additionally, we address the limitations and future outlook of AI in coding. By offering actionable insights, this framework helps to guide researchers in effectively leveraging AI to enhance coding practices and education, accelerating scientific progress.

Accurate time series forecasting, predicting future values based on past data, is crucial for diverse industries. Many current time series methods decompose time series into multiple sub-series, applying different model architectures and training with an end-to-end overall loss for forecasting. However, this raises a question: does this overall loss prioritize the importance of critical sub-series within the decomposition for the better performance? To investigate this, we conduct a study on the impact of overall loss on existing time series methods with sequence decomposition. Our findings reveal that overall loss may introduce bias in model learning, hindering the learning of the prioritization of more significant sub-series and limiting the forecasting performance. To address this, we propose a hybrid loss framework combining the global and component losses. This framework introduces component losses for each sub-series alongside the original overall loss. It employs a dual min-max algorithm to dynamically adjust weights between the overall loss and component losses, and within component losses. This enables the model to achieve better performance of current time series methods by focusing on more critical sub-series while still maintaining a low overall loss. We integrate our loss framework into several time series methods and evaluate the performance on multiple datasets. Results show an average improvement of 0.5-2% over existing methods without any modifications to the model architectures.

Automatically graded programming assignments provide instant feedback to students and significantly reduce manual grading time for instructors. However, creating comprehensive suites of test cases for programming problems within automatic graders can be time-consuming and complex. The effort needed to define test suites may deter some instructors from creating additional problems or lead to inadequate test coverage, potentially resulting in misleading feedback on student solutions. Such limitations may reduce student access to the well-documented benefits of timely feedback when learning programming. In this work, we evaluate the effectiveness of using Large Language Models (LLMs), as part of a larger workflow, to automatically generate test suites for CS1-level programming problems. Each problem's statement and reference solution are provided to GPT-4 to produce a test suite that can be used by an autograder. We evaluate our proposed approach using a sample of 26 problems, and more than 25,000 attempted solutions to those problems, submitted by students in an introductory programming course. We compare the performance of the LLM-generated test suites against the instructor-created test suites for each problem. Our findings reveal that LLM-generated test suites can correctly identify most valid solutions, and for most problems are at least as comprehensive as the instructor test suites. Additionally, the LLM-generated test suites exposed ambiguities in some problem statements, underscoring their potential to improve both autograding and instructional design.

Large Language Models (LLMs) are widely adopted for assisting in software development tasks, yet their performance evaluations have narrowly focused on the functional correctness of generated code. Human programmers, however, require LLM-generated code to be not only correct but also optimally efficient. We propose PerfCodeGen, a training-free framework that enhances the performance of LLM-generated code by incorporating feedback based on runtime during test case execution into the self-refinement iterations. With PerfCodeGen, we achieve speedups for a significantly higher proportion of problems compared to using the base LLM with sophisticated prompting techniques. Applied to open language models like Phi-3-mini, PerfCodeGen achieves runtime efficiency comparable to prompting powerful closed models like GPT-4. We achieve state-of-the-art runtime efficiency on benchmarks such as HumanEval, MBPP, and APPS, frequently surpassing the ground truth reference solutions with PerfCodeGen using GPT-3.5 and GPT-4. Additionally, we demonstrate the effectiveness of our approach in enhancing code quality across a range of open LLMs of varying sizes including Phi-3-mini, Llama 3 8B, Mixtral 8x7B, Command R, and Llama 3 70B.